Research map / Detection and segmentation / CNN features
Semantic Segmentation by Early Region Proxy
Yifan Zhang, Bo Pang, Cewu Lu
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Replace regular patch-grid prediction with learnable region proxies and per-region classification | ViT-Ti/16 · ADE20K val | mIoU (SS): 39.0 -> 42.1 | Improved |
| Replace regular patch-grid prediction with learnable region proxies and per-region classification | ViT-S/16 · Cityscapes val | mIoU (SS): 76.1 -> 75.9 | Mixed |
| Use 3–5 early Transformer layers for token/region feature extraction rather than M=0 (ablation) | RegProxy-S/16 · ADE20K | mIoU: 46.3 -> 47.2 | Improved |
| Use 3–5 early Transformer layers for token/region feature extraction rather than M=0 (ablation) | RegProxy-S/16 · Cityscapes | mIoU: 76.8 -> 79.0 | Improved |
| Make the token head too deep (M=12), leaving no later layers for region-context encoding (ablation) | RegProxy-S/16 · ADE20K | mIoU: 47.2 -> 45.7 | Hurt |
| Increase output resolution from 1×1 to 4×4 subcells (ablation) | RegProxy-Ti/16 · ADE20K | mIoU: 39.30 -> 40.76 | Improved |
| Increase output resolution from 1×1 to 4×4 subcells (ablation) | RegProxy-Ti/16 · Cityscapes | mIoU: 73.11 -> 75.08 | Improved |
| Add a 3×3 depth-wise convolution to the affinity head (ablation) | RegProxy-Ti/16 · ADE20K | mIoU: 40.01 -> 40.72 | Improved |
| Add a 3×3 depth-wise convolution to the affinity head (ablation) | RegProxy-Ti/16 · Cityscapes | mIoU: 75.04 -> 74.96 | No clear effect |
| Initialize with ImageNet-21k pretraining rather than random initialization (ablation) | RegProxy-Ti/16 · ADE20K | mIoU: 13.96 -> 40.76 | Improved |
| Initialize with ImageNet-21k pretraining rather than random initialization (ablation) | RegProxy-S/16 · Cityscapes | mIoU: 49.36 -> 78.71 | Improved |
| Add explicit region-semantic homogeneity regularization (ablation) | RegProxy-Ti/16 · ADE20K | mIoU: 40.76 -> 40.66 | Hurt |
Closest papers by meaning
- Segmenter: Transformer for Semantic Segmentation (ICCV 2021 · similarity 0.89)
- Rethinking Semantic Segmentation From a Sequence-to-Sequence Perspective With Transformers (CVPR 2021 · similarity 0.88)
- Rethinking Decoders for Transformer-based Semantic Segmentation: A Compression Perspective (NeurIPS 2024 · similarity 0.88)
- REN: Fast and Efficient Region Encodings from Patch-Based Image Encoders (NeurIPS 2025 · similarity 0.87)
- Geometry-Guided Semantic Reconstruction for 3D Instance Segmentation (NeurIPS 2026 · similarity 0.87)
- Region-Based Representations Revisited (CVPR 2024 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
